Post #1887127
2026-05-02 09:36 UTC
Replies (5)
-
@frumble@chaos.social 2026-05-02 10:58
@festal Maybe, but according to this, 'frontier' models are just a year ahead of the freeware models. Might change but it seems totally possible some firms will invest in competitive training the road ahead even when the VC bubble finally bursts. https://chaos.social/@frumble/116484941291574137
-
@DrSeltsam@digitalcourage.social 2026-05-02 11:54
@festal good programmers are also only available to corporate clients
-
@lalonsander@chaos.social 2026-05-02 15:53
@festal I think, on the contrary, LLMs are an area in which advanced tech has been available to the public at the same time as or even before it was available to, for example, the military. There are already providers offering low cost inference for quite advanced open source models. Those, quite certainly, are not being subsidized. I also doubt newer models are/will be significantly more powerful. A lot of power of current LLMs comes from the software built around the AI inference.
-
@d1@autistics.life 2026-05-03 18:48
@festal I totally agree - any LLM which is even a notch or two less than the best is effectively worthless.
-
@alienocene@mastodon.social 2026-05-04 06:04
@festal But how, then, would this help transform / improve a degraded informational milieu and prevent the cognitive collapse of AI? For expensive models would still face the shitty products of cheap AI. This brings a “Maxwell’s demon”-type problem...